Fig. 3
From: kalis: a modern implementation of the Li & Stephens model for local ancestry inference in R
Log-log plot of milliseconds per variant performance (y-axis) of the forward algorithm on 10,000 haplotypes, against the number of variants propagated (x-axis). Each panel is a different loop unrolling depth (panel title gives loop unrolling level). Line colour denotes number of CPU threads, whilst a dashed line indicates vanilla C and a solid line indicates hand-coded AVX2 instructions. In total, using AVX2, 48 threads, and loop unrolls to depth 8, it takes less than 10 seconds to propagate a \(10000 \times 10000\) forward table over 10,000 variants